A multi-layer echo state network for efficient DDoS detection in resource-constrained environments

Abstract

This study proposes a multi-layer Echo State Network (ESN) model for effectively detecting DDoS attacks on resource-constrained and low-memory devices. Generally, these low-memory devices, common in smart homes, healthcare, and industrial applications, do not have enough computational resources to run traditional deep learning methods of DDoS attack detection. This makes the devices much more vulnerable to attacks. While previous works have focused mainly on improving detection accuracy, they have failed to consider vital trade-offs between resource utilization and detection performance. The proposed ESN model achieves 99.33% and 99.99% accuracy in CICDDoS2019 and CICIoT2023 datasets respectively. With only 640 trainable parameters, it ensures high performance with minimum consumption of computational resources. The proposed model has 1.27% and 0.06% CPU utilization in CICDDoS2019 and CICIoT2023. The CPU utilization is much lesser compared to LSTM, RNN, CNN, and state-of-the-art models, respectively. This makes our model a lightweight architecture suitable for devices with limited memory and processing power. The paper presents an efficient, lightweight model for the security of low-resource environments and robust DDoS detection without loss of accuracy.